Unaddressed Problems
Microphone array signal processing is a technical domain where traditional speech and array processing meet. The primary goal is to enhance and extract information carried by acoustic waves received by a number of microphones at different positions. Due t
- PDF / 68,445 Bytes
- 6 Pages / 439.37 x 666.142 pts Page_size
- 28 Downloads / 146 Views
10.1 Introduction Microphone array signal processing is a technical domain where traditional speech and array processing meet. The primary goal is to enhance and extract information carried by acoustic waves received by a number of microphones at different positions. Due to the random, broadband, and nonstationary essence of speech and the presence of room reverberation, microphone array signal processing is not only a very broad but also a very complicated topic. Most, if not all, of the array signal processing algorithms need to be re-developed and specially tailored for the problems with the use of microphone arrays. Therefore, one cannot expect that one book could and should cover all these problems. As a matter of fact, in this area and every year, a great number of Ph.D. dissertations and numerous journal papers are produced. A key thing that we want to demonstrate to the readers is how an algorithm can be developed to properly process broadband speech signals no matter whether they propagate from far-field or near-field sources. We selected those problems for which we achieved promising results in our research as examples. But the unaddressed problems in microphone array signal processing are also important. They are either still open for research or better discussed in other books. In the following, we will briefly describe the state of the art of three unaddressed problems and provide useful references to help the reader for further detailed studies.
10.2 Speech Source Number Estimation Microphone arrays, as a branch of array signal processing, offer an effective approach to extending the sense of hearing of human beings. Genetically, enhancement of acoustic signals from desired sound sources and separation of an interested acoustic signal (either speech or non-speech audio) from other competing interference are the primary goals. But whether these goals can be
218
10 Unaddressed Problems
satisfactorily achieved depends not only on speech enhancement and separation algorithms themselves (as evidenced by the great efforts made in most, if not all, parts of this book for their advancement), but also on an array’s capability of characterizing its surrounding acoustic environment. Such a characterization, sometimes termed as auditory scene analysis (ASA), includes determination of the number of active sound sources, localization and tracking of these sources, and the like technologies. Speech source number estimation is an important problem since many of the algorithms for processing microphone array signals make the assumption that the number of sources is known a priori and may give misleading results if the wrong number of sources is used. A good example is the failure of a blind source separation algorithm when the wrong number of sources is assumed. While acoustic source localization and tracking have been discussed in the previous chapter, no section was devoted to the problem of speech source number estimation in this book. This is not because speech source number estimation is easy to solve, but on the c
Data Loading...
